Log in

Coach McGuirk

allogenes in statisticians

Textbook Topics

I am currently working on an introductory textbook and some additional materials (including textbook/class materials for an intermediate course). I have a question for you.

Set aside any preconceptions about what must be in an introductory (or intermediate) book on statistics. What should be in such a book? Introductory stats books still suffer from being very 1950's oriented in terms of topics, emphasis, lack of computing, etc. (If you think computing should still play no role in such a book, let me know that, too.) Statistics has changed a lot over the decades, should not the books change, too?

If you are a statistician, what is important that is missing? If you are from another field, what do you find missing or what should not be there? Please let me know if you are a statistician or not, too, that would help.

Any input would be great, and be really appreciated! Thanks!

Crossposted at stat_geeks


Hi! I have a general comment that is format-related and applicable to both introductory and more advanced levels. I would like to emphasize the importance of linking the content of the course to the hands-on tutorials on the use of statistical software and case studies that describe the practical application of these concepts. In my graduate school career I rarely gave much attention to heavily theoretical textbooks, but I still keep more applied ones as a reference. I hope that helps! -Marat
Thanks! I feel the same way. I can see how one integrates case-studies easily, and that is definitely part of this project. All the people working on it agree on that.

If you have any examples--how have you seen software integrated in an especially good way? I guess my concern with software integration is that big publishers can make supplements for textbooks (so they can be changed as software evolves or for different packages) but this is for a smaller publisher and that is not possible. Any material will have to be in a single volume. Do you think it works better fully integrated in text or as separate chapters/sections?

This may be a side point, but one goal for this project is to make low-cost textbooks that do not get new editions every single year. That raises costs to students who can't find used copies. (Which, of course, is why the bigger publishers do it.)

I did not mean that you actually need to sell a copy of a statistical computer program with each book. I just meant that you need to teach students how to use the software that they already have access to. If your audience does not have access to such software, you can always focus on Excel or free software like R.

This book may serve as a good example:

Each chapter has three main components: a) description of a statistical concept; b) a case study, where this statistical method is utilized to solve a real-life problem - this section also contains a partial write-up of the results in a proper academic journal format; c) syntax (code) for SAS and SPSS, that can be used to conduct the described analyses.

As for your last point - I think that neither statistical methods nor computer code used for statistical computing change very often, so there is no real need to publish a new edition every year... unless you want to rip-off some rich college kids))
I didn't mean include the software, just have a separate book for the instructions---like some of the older books I have taught from which have supplements for stata, spss, and minitab. We need to embed everything in one text for this project.

And yeah, we don't expect a lot of changes every year, but we're trying to work on a line of books that old timers like me remember: ones that basically have one or two editions in total and last for a decade. Over the last decade software has changed. We're leaning toward using R as a core, and building a specific package to implement the text.

And yes; we're avoiding ripping students off. :-)
I would say discussions of how to handle missing data.

also, I have not seen any intro. textbooks that cover linear regression from a Bayesian framework. we always assume a classical one -- with the response that we don't want to confuse the students. My questions is why not introduce them to both frameworks and let them decide which is better for a particular problem.

also, emminant statisticians acknowledge problems with our introductory discussions with type I and type II error. i.e. what we say on the subject is not fully accurate. I think that someone should go back to the current statistical literature and correct this.

Why not introduce "advanced" computational topics at a simple level i.e. bootstraping.

if you are creating suplementary material as well I would say simple material on how to use R. most books on the topic are too sophisticated for an introductory user or are out-of-date or both.

last, but not least, better problems for the students to work through. some of the current problems are silly and are from the era of calculators and hand-calculations.

p.s. I am a statistician and a graduate student in statistics. anyways, HTH
Thanks for all the ideas!

I am tending toward introducing the bootstrap and randomization tests early on to give a good general tool, and then introduce other methods later. The "chemically pure" Bayesians I have spoken to, however, hate that idea. One put it this way: "bootstrapping is frequentist statistics run amok." So there does appear to be an issue there. Not sure how to deal with those sorts of issues...
simple, realize that you can't please evryone. you can note that strict bayesians, and note that Bayesian statistics is a philosophy / school of statistics ideaology, may disagree with doing this. many disciplines have different schools of thought associated with them. some people within these various schools may be dogmatic.

also for an interesting and slightly more theoretical, at least by North American standards, 1st year statistics text book I suggest looking at Jeffery Rosenthal's book.

btw, apearently it is possible to have a Bayesian bootstrap. I personally don't know much about how it works.

I would also get rid of the silly standrd normal tables from teh 50's that everyone seems to put in their books.

these days very few people, at least in north American and i would suspect Europe do not have access to a computer, often a laptop.

also you need to decide and that would help with your content to what level you want to pitch the book i.e. to a business stats course, psychological stats course, or a regular biology / theoretical stats course.


Well, in essence that is the main point of my problem: I want a sense of what other people think is the right foundations for a modern introduction. Should such a book be Bayesian at root? Why or why not? I have surveyed a large number of relatively recent books and they are all clearly committed to frequentist statistics. But most statisticians agree that that is "incoherent" logically. Then they go and teach it that way anyway. Maybe it is time to make the first book in stats actually match what the professors later on expect to teach.

I know from some survey work we've done, that a book based on bootstraps will have problems from the teachers; but a Bayesian book has different sorts of problems.

So I guess my real question is "what orientation do people want to see in such a book?" It sounds to me like you would prefer something eclectic and not married to real foundational issues. That seems to be the preference of applied statisticians I have surveyed. Oddly, social science people want a more Bayesian orientation but then get upset when probability notation/math shows up (go figure!). My theoretical statistician contacts appear to prefer a strong Bayesian foundation and less computation. As it is a general intro book, with a data focus, that is not going to happen. :-)

And yeah, computation will be embedded and tables removed. I'll look up the Rosenthal book. Thanks!
Throw in a sidebar/chapter titled "Oops" which shows how the technique discussed in the chapter can give misleading results (you could always cite research papers if you don't want to make up an example) -- there are many ways in which statistics can mislead, and having that nagging awareness may lead to more intelligent use of the calculations.
That is a good idea! I know of one book that did something similar, I think it was Norman and Streiner's intro biostatistics book. Thanks!

November 2011

Powered by LiveJournal.com